Publication - OKB5azEUNHSRyl0fzQNcRd0ZsYD2

The article is titled “Our AI model can help improve indoor ventilation during wildfire season.” It was originally published on September 30, 2025 and is authored by Hoda Khalil, Adjunct Research Professor and Lecturer, Systems and Computer Engineering, Carleton University and Gabriel Wainer, Professor, Systems and Computer Engineering, Carleton University.

A recent report from the University of Chicago’s Air Quality Life Index found that wildfires are worsening air quality in Canada. The report found that in 2023, wildfires caused concentrations of particulate matter to rise to levels not seen since the index started taking records in 1998.
This summer, Canada experienced one of the worst wildfire seasons on record. Fires caused thousands to evacuate their homes and smoke periodically blanketed cities, causing outdoor air quality to deteriorate.
When we smell or see smoke, the first thing many of us might think to do is close our windows. However, wildfire smoke contains small fine particulate matter (PM2.5) that can pass through small openings or gaps.
In 2023, wildfires in Canada caused more greenhouse gas emissions than all other sources combined. That means designing safer indoor spaces is a public health imperative. But how can we develop indoor spaces that are well-ventilated and safe from the harmful effects of smoke?
Enhancing indoor air quality
Answering this question would traditionally require going through a real-world process of trial and error in various spaces. Such a process is time-consuming and not always feasible. However, we recently developed a framework integrating modelling and simulation with deep learning techniques to help answer this question.
We know that enhancing indoor air quality, whether through improved ventilation, an optimal occupancy-to-area ratio or other room setting adjustments, can improve health and reduce the spread of infections.
The next step for researchers and designers is to determine the best indoor design features to reduce carbon dioxide concentration. Such features include rooms dimensions, the location of ventilation ports, ventilation levels, where windows are, maximum number of occupants, seating arrangements and so on.
How our model works
Our framework tackles two pertinent problems: the lack of verified, accurate information and the inefficiency of producing and studying simulation results for many combinations of settings.
We use an advanced mathematical model and associated software tools that allow us to simulate varied enclosed spaces with different settings, and to collect simulation results.
The simulated data is then further used to form a data set to train an AI algorithm — in this case, using a deep neural network. Designers can use the trained network to predict unknown settings of the closed space when other settings are altered.
The framework allows designers to simulate how changes in room layout, such as the number vents and where they are placed, or the density of occupants, could impact well-being. For example, the framework can estimate how many people might get sick in a given space, helping architects and planners adjust configurations to minimize infection risk before construction begins.
We used several case studies from university laboratory settings to validate the framework. In one case study, our research team could create 600 simulation scenarios of different laboratory designs. The simulation results produced a rich dataset that would be nearly impossible to replicate in real life due to cost and logistical constraints.
The resulting dataset is used to train a machine learning algorithm to predict where and how many people might be exposed to high levels of carbon dioxide. With that information in hand, it’s easier to make smart decisions about where to place ventilation ports or how many people should safely occupy a room under specific conditions.
Future studies needed
Across Canada, researchers are leveraging machine learning to study indoor air quality in homes, schools and offices. Our findings suggest that this approach is well-suited for studying how carbon dioxide spreads in indoor environments.
However, broader study is still needed. To date, case studies have focused exclusively on a university environment. Yet our framework is designed to be scaleable and adaptable to wide range of indoor spaces. Future research should expand to schools, gymnasiums and residential buildings to strengthen the trust in the framework and refine its predictive power.
As climate change intensifies wildfire seasons, Canadians will spend more time indoors avoiding smoke. The good news is that we have the tools, data and the scientific insight to make indoor spaces healthier and safer for everyone.
We may not have the means to control the air outside, but we can design our spaces to control the quality of the air inside.

Our AI model can help improve indoor ventilation during wildfire season

September 30, 2025 • My Project Name

Read Article

About My Project Name

https://www.mygroup.com